Datadog Code Security MCP collects usage telemetry to help the team understand how the tool is used and improve it over time.
Please see the Datadog Privacy Policy for more information.
Every scan emits one telemetry event per scan type plus, when multiple types are run together (scan all / datadog_code_security_scan), an additional aggregate event that covers the whole batch. This lets you count executions of any single type (e.g. iac_scan) directly by operation, without inferring counts from the aggregate.
| Invocation | Events emitted |
|---|---|
Single-type (scan sast, datadog_sast_scan) |
One per-scan event (operation=sast_scan, standalone=true) |
Multi-type (scan all, datadog_code_security_scan) |
One aggregate event (operation=code_security_scan) plus one per-scan event per executed type (standalone=false) |
| Initialization failure (binary missing, bad path) | One aggregate event only (success=false) — no per-scan events since nothing executed |
Other CLI command (setup, version, generate-sbom) |
One operation event for the command |
Ephemeral SAST rule (ephemeral-scan, datadog_ephemeral_sast_scan) |
One operation event (operation=ephemeral_sast_scan, auth_method=none) with findings_count |
Custom rule save (promote-sast-rule, datadog_promote_sast_rule) |
One operation event (operation=promote_sast_rule) with auth_method |
Common to all events:
| Field | Description |
|---|---|
operation |
Stable operation name (e.g. sast_scan, code_security_scan, generate_sbom, setup, ephemeral_sast_scan, promote_sast_rule) |
interface |
cli or mcp |
duration_ms |
Wall-clock time for this event: per-scan time for per-scan events; total elapsed time for aggregate |
success |
Whether the invocation succeeded |
error.kind |
Categorized error type (BinaryNotFound, AuthRequired, InvalidArguments, PathNotFound, Timeout, Network, ScanError, GitError, SetupError, Unknown) |
error.message |
Short curated description for Error Tracking: a per-kind default (e.g. scan execution failed) or a more specific string for known sub-cases (optionally suffixed with the process exit code). Hardcoded and path-free — never the raw error text |
os, arch, go_version |
Runtime platform info |
ci |
Whether a CI environment variable is set |
session_id |
Random UUID stable for one MCP server lifetime (MCP mode only) |
usr.id |
Anonymous install ID — random UUID unique per installation |
version |
CLI version |
env |
Deployment environment (development or production) |
team |
Fixed telemetry ownership tag (k9-iac) |
Depending on the operation:
| Field | Description |
|---|---|
caller |
skill when a Datadog agent skill invokes the CLI with --called-by-skill or an MCP tool with called_by_skill: true; omitted for direct CLI and for MCP calls that omit the arg |
auth_method |
Authentication source category (none, env_var, or auth_provider); never credentials |
binary_versions |
Installed scanner version/status by scanner (X.Y.Z, not_found, or unknown) |
used_binary_versions |
Installed version/status of only the binaries this scan type or operation actually invokes (e.g. {"static_analyzer":"1.2.3"} for SAST; {"sbom_generator":"…","security_cli":"…"} for SCA; {"sbom_generator":"…"} for generate_sbom). A scoped subset of binary_versions — use this for per-binary version distributions to avoid double-counting the full inventory carried on every event. Present on per-scan events and on the generate_sbom operation event |
first_run |
Whether this is the installation's first recorded invocation |
is_git_repo |
Whether the target is inside a Git repository |
is_worktree |
Whether the target is inside a Git worktree |
auth_configured |
Whether the MCP server started with authentication configured |
libraries_count |
Number of package coordinates submitted to a library scan |
detailed |
Whether the non-sensitive version --detailed display mode was selected |
notice |
Curated message for a non-fatal informational signal from a successful operation (e.g. a zero-component SBOM generation, or a scan that had nothing to check). Present only when surfaced; appears on per-scan events and on generate_sbom. A hardcoded, path-free string — never raw error text |
config_errors |
Category names for telemetry-config initialization failures; never raw error text or paths |
id_ephemeral |
Whether the anonymous install ID could not be persisted |
config_rename_attempts |
Number of atomic-rename attempts the config write needed (1 on a clean write, > 1 under cross-process write contention, seen almost exclusively on Windows). Present on config-persistence events, including the standalone telemetry_config_rename_contended signal emitted when a write succeeded only after retries |
Per-scan events only (sast_scan, secrets_scan, sca_scan, iac_scan):
| Field | Description |
|---|---|
standalone |
true when the scan type was invoked on its own; false when part of a scan all batch |
findings_count |
Number of findings for this scan type — no content or details |
severity_breakdown |
Per-severity finding counts for this type (e.g. {"HIGH":2,"MEDIUM":2}) |
paths_count |
Number of file paths passed to the scan |
output_format |
human or json — CLI events only |
SAST includes LOW findings by default. When the caller selects a higher
minimum severity, findings_count and severity_breakdown reflect the
filtered result. In-source-suppressed results remain excluded.
Aggregate event only (code_security_scan):
| Field | Description |
|---|---|
scan_types |
Comma-separated list of scan types requested (e.g. sast,secrets) |
findings_count |
Total findings across all scan types |
scan_types_breakdown |
Per-detection-type finding counts (e.g. {"sast":3,"secrets":1}) |
severity_breakdown |
Per-severity finding counts across all types (e.g. {"HIGH":2,"MEDIUM":2}) |
scan_durations_breakdown |
Per-scan-type wall time in ms (e.g. {"sast":1200,"secrets":340}) — sourced from goroutine timing |
partial_errors_count |
Number of scan types that failed when others succeeded (graceful degradation) |
partial_errors_breakdown |
Per-scan-type error kind for each failed type (e.g. {"iac":"BinaryNotFound"}) — kinds only, no raw messages |
notices_count |
Number of scan types that emitted a non-fatal notice (e.g. a zero-component SBOM scan) |
notices_breakdown |
Per-detection-type curated notice message for each type that surfaced a notice (e.g. {"sca":"No components detected..."}) — hardcoded, path-free strings only |
batch_id |
Random UUID shared with every per-scan event in this batch — use to isolate one scan all execution |
paths_count |
Number of file paths passed to the scan |
output_format |
human or json — CLI events only |
Tip: To isolate all events from a single
scan allrun, filter bybatch_id:<uuid>. Both the aggregate event and all per-scan events (standalone:false) share the samebatch_id. Single-type standalone scans do not have abatch_id.
- Source code, file contents, or scan findings
- Custom rule source, rule or ruleset names, descriptions, or examples sent by
ephemeral_sast_scanandpromote_sast_rule - Detected secrets or vulnerability details
- Raw error messages (
error.messageis a hardcoded, curated description per error kind for Error Tracking — never the raw error text) - Absolute file paths, directory structure, usernames, or repo paths
- Stack traces
- Machine identifiers or IP addresses (beyond what the server logs server-side)
- Arbitrary command arguments or user-provided flag values; only explicitly
documented coarse modes such as
output_format,detailed, andcaller=skillare collected - Any user-identifiable information
Collected telemetry is stored in a dedicated Datadog index with a 30-day retention period, after which it is deleted.
Any of the following disables telemetry completely:
# 1. Per-invocation flag (CLI)
datadog-code-security-mcp --no-telemetry scan sast ./src
# 2. Environment variable (persists for the shell session)
export DD_CODE_SECURITY_TELEMETRY_DISABLED=1
# 3. DO_NOT_TRACK standard (https://consoledonottrack.com/)
export DO_NOT_TRACK=1
# 4. Config file — set telemetry_enabled to false in:
# ~/.datadog-code-security-mcp/config.json
# { "telemetry_enabled": false }Telemetry is also automatically disabled if the compiled-in client token is absent (e.g. in local dev builds built without TELEMETRY_CLIENT_TOKEN).
The MCP server (start) checks whether a newer release of
datadog-code-security-mcp is available, so the agent can tell the user how to
upgrade and refresh the skills. This is separate from telemetry and sends no
telemetry fields:
- At most once a day, in the background, the server sends an unauthenticated
GETto the GitHub API (https://api.github.com/repos/datadog-labs/datadog-code-security-mcp/releases/latest). The request carries only aUser-Agentofdatadog-code-security-mcp/<version>; GitHub sees the request's IP address, as for any download. - The result is cached in
~/.datadog-code-security-mcp/update-check.json(latest_versionandchecked_at). Server start reads this cache and never waits on the network, so a new release is announced from the next session. - When the cached release is newer than the running one, the server appends one sentence to its MCP initialize instructions naming both versions and the upgrade path. Otherwise the instructions are unchanged.
- Failures (offline, proxy, rate limit) are silent and retried after a day. Development and prerelease builds never check.
Any of the following disables the update check:
export DD_CODE_SECURITY_UPDATE_CHECK_DISABLED=1
export DD_CODE_SECURITY_TELEMETRY_DISABLED=1
export DO_NOT_TRACK=1On the first invocation, a short notice is printed to stderr explaining what is collected and how to turn off data collection. It is shown once and then suppressed for subsequent runs (controlled by first_run_notice_shown in ~/.datadog-code-security-mcp/config.json).
In MCP server mode, stderr is captured by the AI assistant client (Claude Desktop, Cursor, etc.) and not directly visible to the user. The primary disclosure for MCP users is the README telemetry section and this document. Running
setupfrom a terminal makes the standard one-time stderr notice visible;setupis non-interactive and does not introduce a separate consent prompt.
The anonymous install ID is a random UUID (uuid.New()) generated once on first run and persisted to ~/.datadog-code-security-mcp/config.json. It is never regenerated, ensuring consistent unique-user counting. It is not correlated with your Datadog account, hostname, or any other identifier.
When telemetry is enabled, scanner versions are collected concurrently once per process. The results are held in memory only and are not persisted.
- Endpoint:
https://browser-http-intake.logs.datadoghq.com/api/v2/logs - Source tag:
ddsource=datadog-code-security-mcp - Auth: client token (RUM-style, publicly embeddable) injected at build time via
TELEMETRY_CLIENT_TOKENCI secret, never yourDD_API_KEYorDD_APP_KEY - Transport: HTTP POST,
Content-Type: application/json, 2 s per-request timeout; process waits up to 3 s at exit for any in-flight POST to complete - Fire-and-forget: telemetry errors are swallowed and never affect scan results
- No stdout writes: telemetry is only written to the intake, never to stdout (which would corrupt the MCP STDIO protocol)
Before the first production release, add the following secret to the GitHub repository settings (Settings → Secrets and variables → Actions → New repository secret):
| Secret name | Description |
|---|---|
TELEMETRY_CLIENT_TOKEN |
RUM-style client token from the Datadog org used for telemetry intake. Obtain from the Datadog Logs > Browser SDK configuration UI. |
If the secret is absent when a release tag is pushed, the pipeline emits a warning annotation and continues — the binary is published but telemetry will be silently disabled in that release.
The canonical Go version for the project is declared in go.mod (go 1.25.5). All CI workflows and the release pipeline use go-version-file: 'go.mod' so they stay in sync automatically; to upgrade Go, update go.mod only.
To test telemetry locally without rebuilding:
# Option A: env var fallback (no rebuild needed)
export DD_CODE_SECURITY_TELEMETRY_TOKEN=<your-client-token>
go run ./cmd/datadog-code-security-mcp scan sast ./testdata
# Option B: .env file (picked up automatically by `make build`)
cp .env.example .env
# Edit .env and set TELEMETRY_CLIENT_TOKEN=<your-client-token>
make build
./bin/datadog-code-security-mcp scan sast ./testdataThe .env file is gitignored, never commit it.
To test with a local server instead of the real intake:
// In tests, use newWithBaseURL(opts, yourServer.URL); see internal/telemetry/telemetry_test.go